Convolutional neural network for semantic segmentation of fetal echocardiography based on four-chamber view

Abstract

The acute shortage of trained and experienced sonographers causes the detection of congenital heart defects (CHDs) extremely difficult. In order to minimize this difficulty, an accurate fetal heart segmentation to the early location of such structural heart abnormalities prior to delivery is essential. However, the segmentation process is not an easy task due to the small size of the fetal heart structure. Moreover, the manual task for identifying the standard cardiac planes, primarily based on a four-chamber view, requires a well-trained clinician and experience. In this paper, a CNN method using U-Net architecture was proposed to automate fetal cardiac standard planes segmentation from ultrasound images. A total of 519 fetal cardiac images was obtained from three videos. All data is divided into training and testing data. The testing data consist of 106 slices of the four-chamber segmentation tasks, i.e. atrial septal defect (ASD), ventricular septal defect (VSD), and normal. The segmentation of the post-processing method is needed to enhanced the segmentation result. In this paper, a combination technique with U-Net and Otsu thresholding gives the best performances with 99.48%-pixel accuracy, 96.73% mean accuracy, 94.92% mean intersection over union, and 0.21% segmentation error. In the future, the implementation of Deep Learning in the study of CHDs holds significant potential for identifying novel CHDs in heterogeneous fetal hearts.

Improvised convolutional auto encoder for thyroid nodule image enhancement and segmentation

“…In case of feed forward network, single slope parameter is learned in each layer whereas with this activation function it is possible for each channel in each layer. Moreover, the mathematical definition of this activation function is given as (3).…”

“…Ultrasonography not only differentiate among solid nodules consisting of cystic components, but it also deals with nodule pathology. Moreover, ultrasonography increases more indication of increased malignancy risk which includes microcalcification, extrathyroidal margins, wide nodules and hypoechoic solid nodules and further, it ignores round shape, cystic composition, smooth margins, spongiform appearance, and echogenicity that are associated with benign disease [3], [4]. In general, thyroid segmentation is carried into two directions [5], [6] i.e., thyroid nodule and thyroid gland segmentation [7], [8]; although both are almost same, further can be categorized in four distinctive part [9]- [13] i.e., shape and contour-based, region-based, hybrid approach, machine learning (ML) approach; these four methods are described as: i) Shape and contour-based approach: In this type of approach, shape or boundary is used as the gland or nodule information for segmentation of nodule or gland in ultrasound (US) image; in this case boundaries between nodule and tissue can be blurred due to image artefacts and image contrast.…”

Segmentation of ten fetal heart components with coarse‐to‐fine cascading and dynamic feature powering

“…The related full names and the corresponding abbreviations of fetal heart components and fetal heart diseases are shown in Table 1 . Rachmatullah et al [27] segmented four chambers based on the dataset including ASD, VSD, and normal data. Yang et al [6] also includes data from five diseases including HLHS, TAPVC, PA/IVS, ECD, and FCR.…”

Segmentation of Squamous Columnar Junction on VIA Images using U-Net Architecture

“…The U-Net-based CNN method is one of the semantic segmentation algorithms that can specifically find the detection object accurately. U-Net architecture is proposed as a segmentation method in this study because it has been proven to be able to complete segmentation tasks in the medical field well, including cardiac fetal Akhiar Wista Arum, Siti Nurmaini, Dian Palupi Rini, Patiyus Agustiansyah, Muhammad Naufal Rachmatullah Segmentation of Squamous Columnar Junction on VIA Images using U-Net Architecture segmentation [24], liver variety segmentation [19], and Brain Tumor segmentation [20]. U-Net is an end-to-end fully convolution network type architecture that contains a convolution layer without a fully connected (dense) layer.…”

“…The post-processing stage makes each image pixel is to 0 for the background and 1 for the foreground. The two best post-processing methods [24] were compared to obtain optimal results in SSK segmentation. The post-processing method used is global thresholding (fixed thresholding), and Otsu thresholding.…”